malloc inside atomic transaction

malloc inside atomic transaction

patrick.marlier's picture
patrick.marlier
Mon, 10/11/2010 - 05:35
Hello, I just want to add a remark about the Intel STM compiler. ICC STM 4.0 
../../test/intset/intset.c(173) (col. 5): error: non [[transaction_safe]] function "malloc" called inside [[transaction_safe]] routine
ICC STM 3.0 
../../test/intset/intset.c(173) (col. 5): error: non tm_safe/tm_pure function "malloc" called inside tm_safe routine
icc complains about malloc() function because it is inside a transaction and it is non safe function. But Intel STM Compiler knows how to manage malloc/free inside transaction. So, why does the programmer have to add explicitly pure or safe attribute to it? Moreover even if it is annotate as pure, the malloc is wrapped. 
callq  0x419630
why there is no problem with free() function? Anyway, thanks a lot for this great work! Sincerely, Patrick Marlier
RSS Top
2 posts / 0 new
Last post
For more complete information about compiler optimizations, see our Optimization Notice.
patrick.marlier's picture
patrick.marlier
Wed, 10/27/2010 - 10:10

Sorry, my mistake... Actually version 3.0 and 4.0 react the same way and free has the same problem as malloc. Here an example:

#include 
#include 

[[transaction_safe]]
void *myalloc()
{
     return malloc(16);
}

int main()
{
   void *area;
   __transaction [[atomic]] {
     area = myalloc();
   }
   if (area == NULL)
     return 1;
   __transaction [[atomic]] {
     free(area);
   }
   return 0;
}

malloc.c(8) (col. 5): error: non [[transaction_safe]] function "malloc" 
called inside [[transaction_safe]] routine

In this example, malloc isn't implicit declared safe whereas free is implicit declared safe. But I am wondering if this behavior is correct : malloc/free is implicitly declared safe in an atomic transaction block and not inside a safe function. Thank in advance for the clarification, Patrick Marlier.

Top

Login to leave a comment.